Quality-Relevant Process Monitoring with Concurrent Locality-Preserving Dynamic Latent Variable Method

Area of Science:

• Chemical Engineering
• Process Control
• Data Science

Background:

• Dynamic process data often contains noise and outliers, challenging conventional monitoring models.
• Existing dynamic latent variable models may lack robustness in real-world industrial settings.
• Accurate monitoring of process and quality variables is crucial for industrial efficiency and safety.

Purpose of the Study:

• To propose a novel concurrent locality-preserving dynamic latent variable (CLDLV) method for quality-related dynamic process monitoring.
• To enhance the robustness and accuracy of fault detection and diagnosis in dynamic industrial processes.
• To effectively extract correlations between process and quality variables despite data contamination.

Main Methods:

• Development of a low-rank autoregressive model to handle autocorrelation and cross-correlation in dynamic data.
• Integration of neighborhood structure information into a partial least squares model to reveal essential data structures.
• Employment of concurrent projection of latent structures for monitoring input and output faults affecting quality.

Main Results:

• The proposed CLDLV method demonstrates improved robustness against noise and outliers.
• Effective extraction of correlations between process and quality variables is achieved.
• Successful detection and diagnosis of output-related and input-related process faults impacting quality.

Conclusions:

• The CLDLV method provides a robust and effective approach for dynamic process monitoring and fault analysis.
• The integration of locality preservation and dynamic modeling enhances the capability of latent variable methods.
• Validated on the Tennessee Eastman process and hot strip mill, the CLDLV method shows significant potential for industrial applications.